Page 31 - ITUJournal Future and evolving technologies Volume 2 (2021), Issue 1

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 1

the time‑triggered traf ic but maintains the deterministic simulation setup as well as main parameters and assump‑
nature and timeliness guarantees in a TSN network. Sev‑ tions are given in Section 5 and results are presented in
eral related scheduling re inements that are orthogonal Section 5.2 and Section 5.3. Finally conclusions and fu‑
to the recon iguration studied in this article have been ex‑ ture work are outlined in Section 6.
amined in [26,40,43,53,74,93]. We note for completeness
that multicast for TSN has been studied in [80,92], while 2. BACKGROUND: IEEE 802.1 TIME SENSI‑
our focus is on unicast traf ic. TIVE NETWORKING
This article extends the prior conference paper [60],
which provided a brief preliminary overview of the de‑ 2.1 IEEE 802.1Qbv: Time Aware Shaper (TAS)
centralized and centralized recon iguration models, but
did not provide the speci ic operational details, nor de‑ TAS’s main operation is to schedule critical traf ic streams
tailed performance evaluations. This present journal ar‑ in reserved time‑triggered windows. In order to pre‑
ticle provides the full operational details as well as com‑ vent lower priority traf ic, e.g., BE traf ic, from interfering
prehensive performance evaluations. with the ST transmissions, ST windows are preceded by
a so‑called guard band. TAS is applicable for Ultra‑Low
1.3 Contributions Latency (ULL) requirements but needs to have all time‑
triggered windows synchronized, i.e., all bridges from
We comprehensively evaluate the performance of TAS for sender to receiver must be synchronized in time [79,85].
recon igurations in the hybrid and fully distributed mod‑ TAS utilizes a gate driver mechanism that opens/closes
els with respect to network deployment parameters, such according to a known and agreed upon time schedule for
as the time period for the Gate Control List (GCL) to re‑ each port in a bridge. In particular, the Gate Control List
peat (whereby the duration of one GCL repetition corre‑ (GCL) represents Gate Control Entries (GCEs), i.e., a se‑
sponds to the CT), the gating ratio proportion, i.e., Gate quence of on and off time periods that represent whether
Control Entry (GCE) proportion, to control the delay per‑ a queue is eligible to transmit or not.
ceived at the receiving end, the signaling impact on ST and The frames of a given egress queue are eligible for trans‑
BE classes, and the packet loss rate experienced at the re‑ mission according to the GCL, which is synchronized in
ceiving end. In particular, we make the following contri‑ time through the 802.1AS time reference. Frames are
butions: transmitted according to the GCL/GCE and transmission
selection decisions. Each individual software queue has
i) We design a CNC interface for a TSN network to glob‑
ally manage and con igure TSN streams, including its own transmission selection algorithm, e.g., strict prior‑
admission control and resource reservation. ity queuing. Whereby, a software queue is the queue be‑
fore the NIC hardware queue takes ownership of the cur‑
ii) We integrate the CNC in the control plane with TAS rently forwarded frame in an 802.1 switch. Overall, the
in the data plane to centrally manage and shape traf‑ IEEE 802.1Qbv transmission selection transmits a frame
ic using the CNC as the central processing entity for from a given queue with an open gate if: ( ) The queue
low schedules as more lows are added. contains a frame ready for transmission, ( ) higher pri‑
ority traf ic class queues with an open gate do not have
iii) We modify and test the model to operate in a dis‑ a frame to transmit, and ( ) the frame transmission can
tributed fashion, i.e., the signaling is conducted in‑ be completed before the gate closes for the given queue.
band and the control plane processing is conducted Note that these transmission selection conditions ensure
at the individual distributed switches. that low‑priority traf ic is allowed to start transmission
only if the transmission will be completed by the start of
iv) We evaluate each design approach for a range of
numbers of streams with different TAS parameters. the ST window for high‑priority traf ic. Thus, this trans‑
We show results for admission ratios, network sig‑ mission selection effectively enforces a “guard band” to
naling overhead, and QoS metrics. prevent low‑priority traf ic from interfering with high‑
priority traf ic [30].
1.4 Organization
2.2 IEEE 802.1Qcc: centralized management
This article is organized as follows. Section 2 provides and con iguration
background information and an overview of related work
on the 802.1 TSN standardization, focusing on the en‑ IEEE 802.1Qcc [3] provides a set of tools to globally man‑
hancements to ST as well as centralized management and age and control the network. In particular, IEEE 802.1Qcc
con iguration. Section 3 shows the complete top‑down enhances the existing Stream Reservation Protocol (SRP)
design of the CNC (hybrid model) and the main com‑ with a User Network Interface (UNI) which is supple‑
ponents that achieve ultra‑low latencies and guaranteed mented by a Centralized Network Con iguration (CNC)
QoS for a multitude of ST streams. Similarly, Section 4 node. The UNI provides a common method of requesting
shows the approach used in implementing the decentral‑ layer 2 services. Furthermore, the CNC interacts with the
ized (fully distributed) TAS recon iguration model. The switch UNI to provide a centralized means for perform‑

© International Telecommunication Union, 2021 15